Skip to content

DEM build options: point cloud vs mesh as source, and the interpolation knob

  • Status: unverified
  • Applies to: Metashape Pro 2.x — and PhotoScan 1.x via the same Chunk.buildDem API; tiled-model-as-source available since Metashape 1.5
  • Edition: Pro
  • Diátaxis: how-to
  • Confidence: high
  • Last reviewed: 2026-05-29

Confidence: high. The source-selection guidance is directly attested by Agisoft support with permalinks (2015, 2019). The Chunk.buildDem parameters and the DataSource enum values are introspection-confirmed on Metashape 2.2.

Problem

You have a project with both a dense point cloud and a built mesh — or sometimes just one of them — and you need to build a DEM. The Build DEM dialog asks for a source data type (point cloud, mesh, or tiled model). Which to pick depends on the dataset's scale, the surface geometry, and the downstream use of the DEM. The defaults are usually OK for typical aerial projects but become wrong choices for façades, large areas, and indoor scans.

This article documents the source-selection rule, the interpolation parameter, and the common pitfalls.

Source data: point cloud vs mesh vs tiled model

Three options:

Source Best for Why
Point cloud (default) Typical aerial surveys; large project areas Direct from depth maps; no meshing step; handles arbitrary terrain shape
Mesh Small areas; building façades; custom planar projections Mesh provides smoother local interpolation; mesh-based DEM has fewer pin-point artefacts at building edges
Tiled model (1.5+) Very large areas where a single-block mesh isn't possible Tiled model preserves mesh quality at scale; DEM rasterises one tile at a time

"Usually we recommend to generate DEM from the dense cloud and use this surface as input for the orthomosaic generation, thus the meshing step can be skipped. For very large areas most likely it wouldn't be possible to generate a single-block mesh, therefore the DEM is recommended for large areas covered. Using Mesh surface may be reasonable for custom plane projections (like building facades, for example) or for small areas, when mesh surface produce more accurate results, for example, by the edges of the roofs." — Agisoft support, 2019-03-22, Metashape 1.5 (permalink)

The recommended workflow on aerial projects: build dense point cloud → build DEM from point cloud → build orthomosaic on top of the DEM. The mesh step can be skipped entirely unless you specifically need it for non-DEM purposes.

For very large areas (kilometres of coverage at high resolution), single-block mesh generation runs out of RAM. The tiled-model path becomes the only mesh-based option:

"In the version 1.5 it is possible, however, to generate the tiled model and then build DEM from it. In case you'll be experimenting with the tiled model generation, you can try to compare the orthomosaic outputs based on DEM from dense cloud and from the tiled model." — Agisoft support, same thread

The interpolation parameter

The Interpolation dropdown in Build DEM controls how the DEM fills cells where the source data is sparse:

Mode What it does When to use
Disabled Cells with no source data are NaN / no-data Small, dense projects where every cell should have direct evidence
Enabled (default) Empty cells filled by interpolating from neighbouring cells, but only within the bounding box Most production workflows; balances coverage and fidelity
Extrapolated Empty cells filled using extrapolation that may extend beyond the convex hull of source data Final-product DEMs that must be rectangular without no-data holes; risk of edge artefacts

"If you are using enabled interpolation, then DEM will be interpolated in the areas that are visible from at least one camera inside bounding box." — Agisoft support, 2015-12-29, PhotoScan 1.2 (permalink)

The Enabled mode's "visible from at least one camera" criterion is important: cells outside any camera's view get no value even with interpolation enabled. This is correct for aerial surveys (the DEM should not invent terrain in unseen areas) but can produce unexpected holes near project boundaries.

For mesh-based DEM generation specifically, the Disabled interpolation option became available only after PhotoScan 1.2:

"Disabled interpolation option for DEM based on Mesh generation will be available in the next version update." — Agisoft support, 2016-01-08, PhotoScan 1.2 (permalink)

If you're on a pre-1.3 version, mesh-source DEMs always interpolate; only point-cloud-source DEMs allow Disabled.

Recipe — DEM from point cloud (default)

Demo verified: ✗ — pending Tier 3 reproduction on a real Metashape install.

import Metashape

chunk = Metashape.app.document.chunk

if not chunk.point_cloud:
    raise RuntimeError("Build the point cloud first via Workflow → Build Point Cloud")

chunk.buildDem(
    source_data=Metashape.DataSource.PointCloudData,
    interpolation=Metashape.Interpolation.EnabledInterpolation,
    # Optional: limit to a region or shape
    # region=chunk.region,
)

print(f"DEM built: {chunk.elevation}")

Recipe — DEM from mesh (for façades or small areas)

Demo verified: ✗ — pending Tier 3 reproduction on a real Metashape install.

import Metashape

chunk = Metashape.app.document.chunk

if not chunk.model:
    raise RuntimeError("Build the mesh first via Workflow → Build Mesh")

chunk.buildDem(
    source_data=Metashape.DataSource.ModelData,
    interpolation=Metashape.Interpolation.EnabledInterpolation,
)

For a façade-orthomosaic workflow, set the chunk's projection to a vertical plane first via chunk.crs or OrthoProjection.type = Metashape.OrthoProjection.Type.Planar — see Orthomosaic in a marker-defined planar projection.

Recipe — DEM from tiled model (very large areas)

Demo verified: ✗ — pending Tier 3 reproduction on a real Metashape install.

import Metashape

chunk = Metashape.app.document.chunk

if not chunk.tiled_model:
    raise RuntimeError("Build the tiled model first via Workflow → Build Tiled Model")

chunk.buildDem(
    source_data=Metashape.DataSource.TiledModelData,
    interpolation=Metashape.Interpolation.EnabledInterpolation,
)

The tiled-model-source DEM scales to project sizes where a single-block mesh would exceed RAM (typically >10km² at sub-decimetre resolution).

Caveats

  • Bounding box matters. Build DEM respects chunk.region. If the region is smaller than your point-cloud extent, only the portion inside the region gets a DEM cell. Verify the region covers what you want before running.
  • DEM resolution defaults to "auto". Metashape picks a resolution based on point-cloud density. For consistent multi-epoch comparison, set chunk.buildDem(resolution=<value>) explicitly.
  • The 1.x → 2.x naming. chunk.dense_cloud → chunk.point_cloud; chunk.elevation (the DEM result) unchanged. Old scripts that branch on if chunk.dense_cloud need the rename.
  • Interpolation Disabled produces non-rectangular outputs. The DEM raster's bounding box is still rectangular, but cells without direct evidence are flagged no-data. This can surprise downstream tools that expect a fully-populated raster.
  • Mesh-source DEM at building edges. The mesh's smooth surface at roof edges produces cleaner DEM cells than the point cloud, which has "step" noise at every vertical surface. This is the strongest argument for mesh-source DEM on small dense projects.
  • Tiled-model DEM may show seams between tiles for certain visualisation contrasts. The seams aren't visible in the underlying data; they're a rendering artefact at steep colour ramps. Smooth out by adjusting the colour-ramp range in the GIS tool.

See also

References

  • Metashape Pro User Manual (2.3), ch. 5 Build DEM — covers the GUI dialog and parameter ranges.
  • Metashape Python API Reference (2.3.1): Chunk.buildDem, Chunk.elevation, Chunk.point_cloud, Chunk.model, Chunk.tiled_model, DataSource enum, Interpolation enum.
  • Forum thread, Orthophoto generation: source from DEM or Mesh?, 2019 — canonical Q&A on source selection; tiled-model option introduction.
  • Forum thread, DEM from Mesh not working, 2015-2016 — interpolation semantics; the Disabled-on-mesh-source release timing.